how secrecy can distort data

http://www.newyorker.com/online/blogs/elements/2013/06/the-problem-with-secret-information.html



The United States government classifies vast amounts of information as “secret.” It then gives millions of people clearance to see what’s hidden from the rest of us. This overclassification is widely acknowledged as bad policy. It clogs up the work of government; with so many needing security clearances, it increases the risk of someone revealing some truly consequential information; and it interferes with the right of a free people to know what their officials are doing. So why does the government work this way? Part of the reason may be simple psychology, according to a new paper, soon to be published in the journal Political Psychology. In three experiments, information was assigned a higher value when people were told it was secret than when the same information was characterized as public.

“When we have a hard problem, we rely on easy-to-understand rules of thumb instead of grappling with it,” said Mark Travers, a social psychologist at the University of Colorado at Boulder, who wrote the paper with his colleagues Leaf Van Boven and Charles Judd. “Information quality is a hard problem, so we have a proclivity to fall back on secrecy as an indicator of quality. And sometimes that works for us. But sometimes it doesn’t”—for instance, when people are too eager to classify documents because that makes them seem more important. Or when people want clearances because of an unquestioned intuition that the secret stuff matters more.

For their first experiment, Travers, Van Boven, and Judd recruited ninety-seven Americans on Amazon’s Mechanical Turk site and asked them to evaluate four foreign politicians seeking office in their respective countries. The participants each played the role of a State Department advisor who had been asked for a recommendation about whether the U.S. should support each politician’s campaign. They were given exactly one piece of positive information and one piece of negative information about each politician. (For example: he handled a border dispute well, but, alas, he appears to be a crook.) The researchers told some participants that the derogatory information had been classified, while the positive material was “widely available to the public.” Others were told the opposite. As the researchers expected, the “secret” label made a difference. Evaluating whether to recommend each candidate on a scale of one (strongly disagree) to five (strongly agree), the participants who thought the negative information was secret settled on an average score of 2.53. Those who thought the positive information was secret delivered an average rating of 3.01. Logically, of course, the classification of the document before them did not have anything to do with the quality of the information it contained.

In a second experiment, the researchers asked fifty-six undergraduate students at the University of Colorado at Boulder to read and evaluate policy papers—one from the State Department and one from the National Security Council—about the sale of military aircraft from Belarus to Peru in the nineteen-nineties. Participants were given both papers (based on real documents from the time that had since been declassified), but only one would have the "declassified" stamp, signalling that it had been secret. Those who thought the National Security Council document was classified as secret gave it, on average, a quality rating of eight out of eleven. Those who thought it had always been public gave it an average grade of 6.89.